Jarrell Hibler

Founder of Premier Marketing. Running a 7-figure agency while building other businesses with AI agents. Building in public, direct and unfiltered.

[Jarrell Hibler @jarrellhibler]

I'm Jarrell. Founder of Premier Marketing, a seven-figure agency with a 35-person team and 94% client retention.

Now I'm building more companies with AI agents, training for the next level 💪, and moving from Charlotte to Miami.

What I post: what I'm building and what breaks, AI agent setups, training, the move. Direct and unpolished.

If you're a founder, love fitness, or are deep in tech and AI, I'd love to connect 🤝. Reply or DM me, and follow along.

[Jarrell Hibler @jarrellhibler]

Running an agency taught me where things break: the handoff. Sales to kickoff. Kickoff to delivery.

Now I'm building with AI agents, and they break in the exact same spot. One finishes, the next one starts with half the context.

Which handoff breaks most in your business?

[Jarrell Hibler @jarrellhibler]

My agents don't grade their own work.

Every change gets reviewed by a second agent from a different model family before it reaches me. It catches the obvious stuff, so my review is about judgment.

Building got easy. Trusting the output is the actual job now.

[Jarrell Hibler @jarrellhibler]

@itaiwins Same in the agency. The wins aren't flashy: a follow-up that never gets missed, a report that shows up on time. Boring and reliable beats impressive and occasional.

[Jarrell Hibler @jarrellhibler]

@_GeorgeReid_ Same thing happens with agencies. Give us the ad account but not the margins or the promo calendar, and we're optimizing blind. The best clients share the numbers before they ask for results.

[Jarrell Hibler @jarrellhibler]

@ganesh2412 This is why my reviewer agent runs on a different model family than the builder. Same model, same blind spots. Still doesn't replace a real user, though. What's your cheapest way to get one in front of it early?

[Jarrell Hibler @jarrellhibler]

@ganesh2412 Cost per accepted result is the right lens. Retries are where the cheap model quietly gets expensive. I'd add review time too: an answer I have to fix myself isn't cheap either.

[Jarrell Hibler @jarrellhibler]

@thejustinwelsh The easiest version I've found: reply to people's posts with a real question, not a pitch. A few of those turned into actual conversations this week.

[Jarrell Hibler @jarrellhibler]

@acadictive Same model, different harness. Each tool wraps it in its own system prompt, context and tools. It's also why I have a reviewer from a different setup check the work. It catches what the first one missed.

[Jarrell Hibler @jarrellhibler]

@Greg_GL_87 "The hours had nowhere to land" is the real diagnosis. Same with handoffs in an agency: nobody drops the ball on purpose, the process just has no place to put it. Did Felix fix it at time entry or at billing?

[Jarrell Hibler @jarrellhibler]

@Greg_GL_87 The patterned half is exactly what you can hand to a person or an agent. If a person can't follow the doc, an agent can't either. The half that doesn't fit a pattern is the actual job.

[Jarrell Hibler @jarrellhibler]

@stas_sorokin_ @KolinHuang The gap being re-reads, not work, is the useful finding. Makes me want to audit what my agents load before the first prompt. Did you measure how much a heavy MCP list adds per re-read?

[Jarrell Hibler @jarrellhibler]

@MGMurray1 "Whoever builds it should not certify it" is the rule that saves me the most cleanup. Did the two copies ever give the same person conflicting answers before you caught it?

[Jarrell Hibler @jarrellhibler]

@MGMurray1 "An agent can't tell abandoned from forgotten" is the line. I keep a shared change log for my agents, and retired rules are the easiest thing to leave in by accident. Do you prune by date, or by owner?

[Jarrell Hibler @jarrellhibler]

@DYoung_AI Removing the author's voice from the evidence is the right framing. My reviewer agent runs on a different model family for the same reason. Does the risk note ever block a merge, or mostly inform it?

[Jarrell Hibler @jarrellhibler]

@DYoung_AI "If the ticket can't stand alone, it was a vibe" is it. Handoffs are where my agents break too: one finishes, the next starts with half the context. Same spot an agency breaks, sales to kickoff.

[Jarrell Hibler @jarrellhibler]

@DYoung_AI Agree on blast radius. My add: the reviewer runs on a different model family than the builder. Same model, same blind spots. How do you decide which files make the high blast-radius list?

[Jarrell Hibler @jarrellhibler]

@cem_hasoglu In an agency, onboarding and retention end up being the same problem. If a client has to repeat themselves after the sale, retention starts cracking right there. Which of those five was hardest to fix at your agency?

[Jarrell Hibler @jarrellhibler]

@webjuice_ie Same setup in spirit: approve the outcome, not the steps. What happens when a client asks for something that contradicts what they approved last month? That's where my handoffs break.

[Jarrell Hibler @jarrellhibler]

@webjuice_ie I'm with you on #1. I approve outcomes, not every step, but a second agent reviews every change before it gets to me. Fully auto works right up until the first silent mistake reaches a client.

[Jarrell Hibler @jarrellhibler]

@webjuice_ie Agree, and it's the same knowledge that keeps clients around. The catch is it usually lives in a few people's heads. If a person can't follow it, an agent can't either, so writing it down helps both.